Cassandra Performance Tuning - Beginning Apache Cassandra Development - page 215

row and key caches in off-heap memory, which means these objects are serialized on a
non-Java heap. That way deserialization will be much faster.
It is recommended to keep heap size moderate, as heap takes memory only from
system memory. It has been observed that it is better to not keep it more than 6 GB (if
system memory is more than 6 GB). You can fetch maxMemory by using Java
runtime:

Runtime rTime = Runtime.getRuntime();
long maxSize = rTime.maxMemory();

If system memory is less than 2 GB, it is recommended to keep 50% of it.
Cassandra requires that the Java Native Access (JNA) library be installed for heap
offloading caches.
To enable the JNA library, you need to download the JNA.jar from ht-
tps://github.com/twall/jna and set on the classpath. That's it!
We can also set off-heap memory allocator within cassandra.yaml :

memory_allocator: NativeAllocator
possible values are NativeAllocator, JEMallocAllocator.

To install JEMallocAllocator, you need to download and install the jemalloc library.
Source for jemalloc is available over GitHub ( https://github.com/jemal-
loc/jemalloc ).

Installing and Configuring jemalloc

The steps to install jemalloc are

1. First clone and download the source from github:

git clone https://github.com/jemalloc/
jemalloc.git

2. Change the directory to jemalloc and run configure :

cd jemalloc
./configure